Model Training System Reducing Teacher Data via Self-Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Supervised learning models require a large amount of data for training, making it desirable to reduce the necessary data quantity for efficient training processes.
Innovation Solution
A model training system that includes a processor configured to input an image, generate a corrected image based on the model's output, and retrain the model using the original image as input data and the corrected image as teacher data, thereby reducing the required teacher data quantity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If supervised learning is performed using traditional methods, then the model can be trained effectively, but a large amount of teacher data is required
Solution Approach 1:
The system enables the model to generate its own teacher data by correcting its own output images. The processor inputs images to the model, corrects the output images, and uses these corrected images as teacher data for retraining, making the system self-sufficient and reducing external data requirements
Solution Approach 2:
The system changes the parameter of teacher data quantity by using a small set of corrected images generated from the model's own outputs instead of requiring large amounts of externally provided teacher data, while maintaining training effectiveness through iterative retraining
Data Source
AI summary
According to one embodiment, a model training system includes a processor. The processor is configured to input a first image to a model and acquire a second image output from the model, and generate a third image by correcting the second image. The processor is configures to train the model by using the first image as input data and using the third image as teacher data.